gcc-10-compat workaround

Copy and fixup of the project https://salsa.debian.org/rpavlik/gcc-10-compat.

Removes some extra steps (delete readme, fixup control file) for building this project on armhf systems with dpkg

Below readme lists a dead repo so the following instructions are how you should use this.

1. clone this repo somewhere on a box with dpkg installed
2. open a terminal and cd into the root gcc-10-compat folder
3. enter the command sudo dpkg-deb --build ./build gcc-10-compat.deb
4. install the package for your architecture with sudo apt install ./gcc-10-compat.deb

** Original ReadMe File From https://salsa.debian.org/rpavlik/gcc-10-compat is below** Maintained at: https://salsa.debian.org/rpavlik/gcc-10-compat

This package exists to fix the buster->bullseye upgrade path, apparently broken by the libgcc1 (and friends) rename (to libgcc-s1, etc) and removal of transitional packages.

The packages from this repo are empty except for the documentation and copyright files: they are only additional "edges" in the apt dependency graph, needed to allow upgrading from Buster to a new enough snapshot of Bullseye that has the libgcc renames. They can be removed once the dist-upgrade/full-upgrade is complete, and deborphan should be able to detect this automatically (they are in section "oldlibs")

I have only tried this for x86_64, and I left out the packages that only applied to e.g. MIPS and HPPA. This should really be fixed in the gcc-10 package, hopefully that will happen before the Bullseye release.

About the problem

There are a number of bugs filed for this - the one I have submitted this info to is https://bugs.debian.org/972820.

This is an apparent bug with the gcc-10 packages in Debian: they renamed several packages and removed the transitional packages after Buster and before the release of Bullseye. The following packages were renamed (to match stated convention/debian policy more closely, I think):

• libgcc1 -> libgcc-s1
• lib64gcc1 -> lib64gcc-s1 (on 32 bit architectures only)
• lib32gcc1 -> lib32gcc-s1 (on 64 bit architectures only)
• libx32gcc1 -> libx32gcc-s1 (on x86_64 only)
• a few others on less common arches

The newly-renamed packages have a Provides: libgcc1 (etc.) property, but it appears that apt is preferring a real (old) libgcc1 package over a virtual package that is only "provide"d by the newer package, since it's trying to not break your system. I am not familiar with the details of the internals of dependency resolution and installation order, but that's how I understand the problem anyway. This repo provides a "real" libgcc1, etc. package to get apt over that hurdle.

In general you know you've hit this bug when you see this on trying to do a dist-upgrade from buster to bullseye (or equivalent in derived distros):

The following packages have unmet dependencies:
 libc6-dev : Breaks: libgcc-8-dev (< 8.4.0-2~) but 8.3.0-6 is to be installed
E: Error, pkgProblemResolver::Resolve generated breaks, this may be caused by held packages.

In some related cases, apt-get will just suggest removal of a number of desired packages that still exist in Bullseye, instead of failing like this. I'm not sure why. Aptitude similarly recommends removing what feels like a large number of desired packages to fix it.

A commonly-suggested "solution" is apt-get install gcc-8-base, which often results in this suggested-mass-removal. Unfortunately, this is the "accepted answer" to most users without the experience to make the missing transitional packages. One reference I found noted to do that before starting the upgrade, which perhaps is enough hint to apt to find a solution in gcc-10? That suggested "solution" doesn't resolve my test case (below), so perhaps there was a previous bug resolvable in this way, with similar symptoms.

I have a pretty minimal reproduction/test case at https://salsa.debian.org/rpavlik/gcc-upgrade-testcase

Instructions

I've pushed this package to the OpenSuSE Build Service OBS instance, so there's a repo you can add which will let you finish upgrading without the unnecessary removals that other tools might suggest. Once the dist-upgrade is complete, please remove the repo and key from your machine: it's no longer needed, and it's good to keep the list of repos you use short.

Before doing this, in case it's needed by the devs working on fixing it in Debian, I'd suggest making a copy of /var/lib/dpkg/status somewhere. It can be very useful in reproducing this or other upgrade bugs.

To use, run these as a normal user with sudo permission:

echo "deb http://download.opensuse.org/repositories/home:/rpavlik:/bullseye-fix/Debian_Testing/ ./" | sudo tee /etc/apt/sources.list.d/bullseye-upgrade-fix.list
curl http://download.opensuse.org/repositories/home:/rpavlik:/bullseye-fix/Debian_Testing/Release.key | sudo tee /etc/apt/trusted.gpg.d/bullseye-upgrade-fix.asc
sudo apt update

After running that, you'll have transitional packages available named libgcc1, etc. that depend on their renamed equivalents in bullseye. This will let your apt-get dist-upgrade or apt full-upgrade complete, or at least get past the problem described above: you'll see

• libgcc-s1 in the list of new packages to install
• libgcc1 in the list of packages that will be upgraded (it's being upgraded to this transitional package)

Once the dist-upgrade completes successfully and you've rebooted, you can remove the repo and key you added above:

sudo rm /etc/apt/sources.list.d/bullseye-upgrade-fix.list
sudo rm /etc/apt/trusted.gpg.d/bullseye-upgrade-fix.asc

If you want to uninstall these empty transitional packages too, I would suggest letting deborphan do it for you.

Hope this helps!

References to this problem

• Debian BTS
  • https://bugs.debian.org/972820
  • https://bugs.debian.org/964477
  • Probably related: https://bugs.debian.org/972936 and its archived relative https://bugs.debian.org/946285
  • This might have been the earlier relative of the bug, referenced in various places: https://bugs.debian.org/954826
• In derivatives
  • https://forums.kali.org/showthread.php?49704-Troubleshooting-Kali-apt-get-upgrade
  • https://forum.sparkylinux.org/index.php?topic=5323.0 - reference is made to a "bug in gcc-8 already fixed in unstable", but the symptoms match the still-existing bug.
• Elsewhere - Note that the list below includes issues filed in incorrect places (e.g. the raspberry pi kernel issue tracker), so don't necessarily add to any of these reports.
  • https://unix.stackexchange.com/questions/592657
  • https://unix.stackexchange.com/questions/602827
  • https://superuser.com/questions/1577742
  • raspberrypi/linux#3812
  • https://www.reddit.com/r/debian/comments/h878ci/fullupgrade_to_debian_testing_fails_due_to/
  • https://www.reddit.com/r/debian/comments/gwzs5n/just_switched_to_testing_from_stable_am_having/
  • possibly related: https://superuser.com/questions/1555536/cannot-solve-the-the-following-packages-have-unmet-dependencies-issue